A Framework for Problem Solving Activities in Multi-Agent Systems

David C. Han, Tse-Hsin Liu, K. Suzanne Barber

Abstract: The basic research issues in multi-agent systems (MAS) include problem decomposition, task distribution, communication, plan synthesis, coordination, conflict resolution, and organization design. For practical implementation, there is a need for an integrated framework that can help MAS designers to select appropriate techniques for building their specific systems. Difficulties in the integration of techniques for each of these issues is due to the interdependencies among the issues themselves. We propose a framework that describes the activities that occur during problem solving. This framework is based upon the premise that meta-level reasoning about the agents’ activities adds flexibility to each agent, allowing them to adjust to changes in their environments or operating conditions. Generally speaking, to accomplish a specific goal, agents must perform a sequence of actions that trigger events and change certain states. A strategy is a decisionmaking mechanism that provides long-term consideration for selecting actions toward specific goals which can help agents to observe the environment, evaluate alternatives, and prescribe and schedule actions. Since different strategies may be more or less appropriate in different situations and within different agent organizations, metalevel reasoning is required for strategy selection. Problem solving activities can be decomposed into three phases: organization design, coordinated planning, and plan execution. Strategies have been developed to address the problems posed by each of these phases. Allowing the agent to select the strategy to apply towards a problem increases flexibility and responsiveness to new situations. For this purpose, each phase can follows the three levels of reasoning described above; reasoning on actions, strategies, and strategic decision making. Coordination during each phase occurs through communication. Upon considering goals the agent wishes to achieve, the agent (or the system designers) must decide the manner in which it will interact with other agents. The result could be static or dynamic organizational structures that are